PCM "sample accuracy" in modern speakers.

[5meohd]

I am a bit confused about "fast" drivers/speakers/woofers etc. I don't think I'm alone there. I lean towards the camp of... its marketing speak.
Someone proposed a decent question since we are both interested in sound-design and mix engineering. "Can X speaker reproduce a 2 sample long transient?". I thought this was actually an interesting way of framing the discussion. For playback/listening 2 samples is completely relative and unnoticed. But for those of us wrangling the complex toolset of limiters, equalizers, compressors, clippers, filters, phase-alignment tools etc... Every sample counts. So, do the current "budget" monitors like JBL LSR305 or even Genelec 8030c have the "speed" to reproduce this detail?

 

[earlevel]

I am a bit confused about "fast" drivers/speakers/woofers etc. I don't think I'm alone there. I lean towards the camp of... its marketing speak.
Someone proposed a decent question since we are both interested in sound-design and mix engineering. "Can X speaker reproduce a 2 sample long transient?". I thought this was actually an interesting way of framing the discussion. For playback/listening 2 samples is completely relative and unnoticed. But for those of use wrangling the complex toolset of limiters, equalizers, compressors, clippers, filters, phase-alignment tools etc... Every sample counts. So, do the current "budget" monitors like JBL LSR305 or even Genelec 8030c have the "speed" to reproduce this detail?

First, speakers give approximations—they do the best they are able to.

In this case, you're asking if they can accurately (a relative term) reproduce something that is not a proper audio signal, but essentially a pair of clicks, of unspecified amplitude, very close to each other (the period of the chosen sample rate)...run through a lowpass filter. Exactly what lowpass filter depends on the hardware.

Then the speakers will do the best they can with that.

But in the end, what has the test told us?

 

[DVDdoug]

Yes, they can reproduce the entire audible frequency range. i.e. A single cycle of 20kHz audio sampled at 44.1kHz will contain 3 samples. The Nyquist limit is half the sample rate. Simplified, that means you need at-least one sample for the positive-half of the waveform and one sample for the negative-half.

We don't hear samples & pulses... We hear (analog) frequency and amplitude. If you play an impulse through an audio system, we perceive the frequency content and the amplitude (and sometimes the duration of any acoustic ringing).

The reconstructed analog audio waveform (which is filtered & smoothed) isn't always "perfect" but our hearing isn't perfect either.

For example, an ideal square wave contains infinite harmonics but above 10kHz you can't hear the harmonics so it doesn't matter if the tweeter can reproduce the harmonics or not.

"Speed" is mostly marketing terminology. Anything that affects transient or impulse response also shows-up in the frequency response.

I don't think you can hear just 2 or 3 samples surrounded by silence but I haven't tried it. It takes a certain time-duration before we can perceive it. And if you can hear it, it would depend on the amplitude. Most real-world "transients" and other high-frequency content isn't very "loud".

If you want to "play around" Audacity can generate "tones" and you could generate a file with one or two samples. Just be careful with longer duration high-frequency test tones. You can fry a tweeter with constant test-tones, and you can even fry it with tones you can't hear!

 

[SIY]

I don't think you can hear just 2 or 3 samples surrounded by silence but I haven't tried it.

That violates Nyquist. The impulse must be bandwidth limited.

 

[NTK]

Expanding on @SIY's reply above, the only bandlimited continuous time signal that will produce a single pulse when sampled is the sinc waveform. If you have two consecutive pulses, the continuous time signal will be the sum of two sinc waveforms, with one delayed by a sampling period.

Top graph is a single pulse at t=0 and its corresponding bandlimited continuous time waveform. Middle graph is a single pulse at t=1 and its corresponding bandlimited continuous time waveform. Bottom graph is the sum of the top and middle graphs, giving the continuous time waveform that correspond to 2 pulses.

If your speaker has the bandwidth to half the sampling frequency, it will be able to reproduce this waveform (but not perfectly due to non-flat frequency response, non-linear distortions, phase shifts, etc.)

 

[earlevel]

I don't think you can hear just 2 or 3 samples surrounded by silence but I haven't tried it.

It's easy to hear even a single sample (of sufficient amplitude). The actual signal you'll be hearing, from the DAC, is a band-limited impulse that lasts much longer, of course. So, it's will sound like a bright, brief snap. A single sample surrounded by silence is something a DSP developer would use to test the impulse response of an algorithm. The spectrum is white (equal energy at all frequencies).

But again, these aren't musical or real world signals (they are valid signals, though). For instance, we need just over two samples per cycle for a sine wave, under half the sample rate. This wouldn't even require an anti-aliasing filter before digitization, at first blush. So you might say, OK, digitize one cycle of such a sine wave. But now you don't have a sine wave, and it's no longer band-limited, so you do need that anti-aliasing filter, which turns the signal into something else before digitization—something that's longer than the three samples or so that you thought you had.

Just mentioning that as a little mind bender. The OP's statement started with samples, two of them. The only generality we can make from that is that the audio output signal will be equivalent to two band-limited impulses. The shape of those impulses will depend on the DAC, but in the case of a linear phase filter, a single sample will result in the classic "pre-ring, post ring" response, and the length of the response will depend on the length of the filter, it's impulse response time (to state the obvious). So, you'd get two of these, overlapped and offset by one sample period. Each essentially looking identical, but with independent amplitude, the amplitude of their sample values.

What I'm saying is you'll get a tiny snap sound, whose exact tone will depend on the exact sample values.

Again, I don't think this is a very useful speaker test. A single sample would be a good test, theoretically, but you'll get exactly the same info from a sweep test, at better s/n.

 

[dc655321]

I don't think you can hear just 2 or 3 samples surrounded by silence but I haven't tried it.

Even one non-zero sample in a sea of silence is audible. See here.
Not gonna win a Grammy though...

 

[Blumlein 88]

Some dacs mute with zeroes. If you try this you may need to combine the impulse with low level noise like -120 db to prevent muting.

 

[5meohd]

It's easy to hear even a single sample (of sufficient amplitude). The actual signal you'll be hearing, from the DAC, is a band-limited impulse that lasts much longer, of course. So, it's will sound like a bright, brief snap. A single sample surrounded by silence is something a DSP developer would use to test the impulse response of an algorithm. The spectrum is white (equal energy at all frequencies).

But again, these aren't musical or real world signals (they are valid signals, though). For instance, we need just over two samples per cycle for a sine wave, under half the sample rate. This wouldn't even require an anti-aliasing filter before digitization, at first blush. So you might say, OK, digitize one cycle of such a sine wave. But now you don't have a sine wave, and it's no longer band-limited, so you do need that anti-aliasing filter, which turns the signal into something else before digitization—something that's longer than the three samples or so that you thought you had.

Just mentioning that as a little mind bender. The OP's statement started with samples, two of them. The only generality we can make from that is that the audio output signal will be equivalent to two band-limited impulses. The shape of those impulses will depend on the DAC, but in the case of a linear phase filter, a single sample will result in the classic "pre-ring, post ring" response, and the length of the response will depend on the length of the filter, it's impulse response time (to state the obvious). So, you'd get two of these, overlapped and offset by one sample period. Each essentially looking identical, but with independent amplitude, the amplitude of their sample values.

What I'm saying is you'll get a tiny snap sound, whose exact tone will depend on the exact sample values.

Again, I don't think this is a very useful speaker test. A single sample would be a good test, theoretically, but you'll get exactly the same info from a sweep test, at better s/n.

This is a very thoughtful and thorough reply.

I think I misrepresented my confusion a little bit. Or, I felt his mention of 2 "samples" was interesting. But in general, if I understand this other parties opinion [which they believe to be fact and are trying to teach others] is that better speakers are able to more accurately reproduce quick changes. So maybe.. a 2 sample click/pop in the midst of an already dense production.

I gave a gut instinct that they are using intuition to understand audio reproduction. Yet, I'm acutely aware of my own ignorance. So.. I seek to improve my own understanding.

This conversation stems from the question of "why can't X learner hear the subtle detail in, say a compressors response/character, when Y instructor is explaining it, but Z learner can". At least this other party admits that it is mostly an ear training thing, then possibly the room, but then they jump to speaker quality and "time axis".

So when I pushed back a little asking about what they mean by time axis... the word "fast" gets brought up.

Again, everyone has had really great responses so far. I'm just a little dim and I still don't have my "answer".

My instinct is that complex program material is actually easier to reproduce for the speaker than square waves. So just because there may be a rhythmically interesting square wave bassline, and lots of neat/tight percussive percussion summed together with delicate harmonic saturation and dynamic range control, it doesn't mean that final signal is more "challenging".

So yes. Some speakers can reproduce a true square wave better than others. But the bottleneck isn't driver material per se.

See... Im ignorant.



Thank you for your time and consideration.

 

[5meohd]

I guess what I'm talking about is exactly this: Transient Response.

So.. whwre does a prosumer studio monitor such as the JBL LSR305 rank in the category of Transient Response.

Is there measurements from NFS tests done here that can show this?

 

[5meohd]

Also, I want to point out that I like this teacher/personality quite a bit.

I just want to be sure he is armed with the best understanding of this.

 

[pjug]

I guess what I'm talking about is exactly this: Transient Response.

So.. whwre does a prosumer studio monitor such as the JBL LSR305 rank in the category of Transient Response.

Is there measurements from NFS tests done here that can show this?

My understanding is if a speaker's frequency response covers (and is relatively flat over) the audio band then it reproduces any transient or step that you would be able to hear. The only question is the degree to which the phase is messed up, and whether that matters. There are threads on ASR that discuss phase. Maybe start with this one:

Does Phase Distortion/Shift Matter in Audio? (no*)

A viewer of my videos and member he suggested that I do a video commenting on a video that Paul McGown did on audibility of phase shifts. Here is Paul's video which was really about a different question (why we need wideband amplifiers) but turned it into phase being an audible problem: Here...

www.audiosciencereview.com

 

[dominikz]

@5meohd What people here are basically saying is that the rise-time of a system that can be approximated as LTI is fully determined by its frequency response - and specifically by its high frequency extension. The mathematical transformation that allows us to convert from time to frequency domain is called Fourier transform - it is what in general allows us to show spectrum of an audio signal, measure frequency response, design EQs etc...
LTI = linear time invariant; note that loudspeakers (and other audio reproduction devices) in general behave as LTI (unless severely overdriven).
Digital audio that is sampled at 44,1kHz contains audio information up to 22,1kHz (in ideal case), so any loudspeaker that plays up to 22,1kHz will be 'fast enough' to reproduce it accurately.

May I refer to this post for some graphical examples and additional basic explanations of how time and frequency responses are related. You may also find this post useful.

Hope this helps!

 

[freemansteve]

I have played with all sorts of speakers from home 'hifi" to guitar/bass systems and small-hall PAs, using 18" speakers down to 6" speakers for LF sounds. I try to hear bass as I like it (ie as it would sound from a bass guitar amp in the case of an err, bass, or real kick drum etc).

I think I get the effect of "fast or slow speakers" as per the title, at least for low frequencies....

We can talk about good things like transient responses, frequency responses etc - all valid and not hard stuff to understand.

I have heard many 'systems' that had convincing bass, and many that did not, but the problem I saw was that while a 18" Fane etc, can go _low_ in the right box, it's usually down to the amp in terms of them sounding "fast", whatever that means. I think you need a ton of power to heave say, an 18" cone back and forth, and unless the amp has a decent damping factor, it may not control it very well, leading to 'farty' bass sounds. I think the bigger the speaker, and the more open the enclosure, the more damping factor you need.

It is possible than all modern amps have great damping factors, but one problem seems to be how it's measured. It is not a simple scalar quantity, because the speaker driver itself and any passive filters all have reactance, (same with the amp's output circuit) hence you are dealing with vector quantities. So the sums are not straightforward, and certainly not a simple ratio of speaker resistance and amplifier internal resistance.

So musing possible causes aside, I'm saying there can be a perception "fastness" or otherwise to some systems at LF at least, and there must be a physical reason for that.

 

[dominikz]

I have played with all sorts of speakers from home 'hifi" to guitar/bass systems and small-hall PAs, using 18" speakers down to 6" speakers for LF sounds. I try to hear bass as I like it (ie as it would sound from a bass guitar amp in the case of an err, bass, or real kick drum etc).

I think I get the effect of "fast or slow speakers" as per the title, at least for low frequencies....

We can talk about good things like transient responses, frequency responses etc - all valid and not hard stuff to understand.

I have heard many 'systems' that had convincing bass, and many that did not, but the problem I saw was that while a 18" Fane etc, can go _low_ in the right box, it's usually down to the amp in terms of them sounding "fast", whatever that means. I think you need a ton of power to heave say, an 18" cone back and forth, and unless the amp has a decent damping factor, it may not control it very well, leading to 'farty' bass sounds. I think the bigger the speaker, and the more open the enclosure, the more damping factor you need.

It is possible than all modern amps have great damping factors, but one problem seems to be how it's measured. It is not a simple scalar quantity, because the speaker driver itself and any passive filters all have reactance, (same with the amp's output circuit) hence you are dealing with vector quantities. So the sums are not straightforward, and certainly not a simple ratio of speaker resistance and amplifier internal resistance.

So musing possible causes aside, I'm saying there can be a perception "fastness" or otherwise to some systems at LF at least, and there must be a physical reason for that.

I agree that often discussed 'fastness' often has to do with bass response (and sometimes treble response).

However I find that amplifier damping factor is usually a relatively small contributor to the issue, even in pretty severe cases; one practical example here (DF=15 amp into 6 Ohm loudspeaker resulting in 1dB FR deviation):

On the other hand room modes and SBIR will often cause 10-20dB deviations in bass response - with perceptual effects ranging from shifted tonality to the extreme 'no bass' or 'one note bass' cases; practical example of setup/room induced one-note-bass here:

So in my opinion (and experience) the perception of loudspeaker 'fastness' is usually connected to relatively obvious frequency response characteristics. E.g. bright (or bass-deficient) loudspeakers, or loudspeakers in setups where room effects cause severe bass nulls, are often perceived as 'fast'; while on the other hand LF room resonance peaks can cause the sound to be perceived as 'slow'.

Luckily IME these kinds of issues are these days very effectively solved by EQ and properly integrated subwoofers

EDIT: Perhaps some will also find this damping factor to FR calculation illustration interesting.

 

[voodooless]

So yes. Some speakers can reproduce a true square wave better than others. But the bottleneck isn't driver material per se.

It’s mostly physical layout of the speaker drivers and the location of the listener (depending on the layout). A non-coaxial multi-way system can be filtered so that it produces a fairly good square wave, but that will only work for a very limited listening area. Outside of it, basically the driver alignment is off, and your perfect square wave is gone.

If you want to cover a larger area, a coaxial helps a great deal. Even better is something like a synergy horn. It has much more and broader control of the wavefront shape and also has the sound sources even closer than an average coax.

 

[5meohd]

It’s mostly physical layout of the speaker drivers and the location of the listener (depending on the layout). A non-coaxial multi-way system can be filtered so that it produces a fairly good square wave, but that will only work for a very limited listening area. Outside of it, basically the driver alignment is off, and your perfect square wave is gone.

If you want to cover a larger area, a coaxial helps a great deal. Even better is something like a synergy horn. It has much more and broader control of the wavefront shape and also has the sound sources even closer than an average coax.

Right. I have owned Danley Labs SH69. So I know this path... I just don't know enough to educate.

Thanks!